Moving objects detection based on tensor ring low rank decomposition.

Area of Science:

• Computer Vision
• Signal Processing
• Machine Learning

Background:

• High-quality camera technology drives demand for efficient video analysis.
• Existing matrix-based methods fragment data and lose spatial information.

Purpose of the Study:

• Propose a novel approach for moving object detection (MOD).
• Enhance video analysis by preserving spatial information and improving efficiency.

Main Methods:

• Combined Low Rank Tensor Ring (TR) decomposition and Tensor Total Variation (TTV) regularization.
• TR decomposition for static background extraction; TTV for moving object representation.
• Utilized low-rank assumption on tensor factors for background and [Formula: see text] regularization for foreground.

Main Results:

• Achieved 3%-8% performance improvement in background separation and f-metric.
• Demonstrated robustness against Gaussian and salt-and-pepper noise.
• Showed suitability for higher-dimensional video processing.

Conclusions:

• The TRLRTTV method offers superior performance in moving object detection.
• The approach effectively handles noise and is applicable to complex video data.
• This novel method advances video analysis techniques beyond traditional matrix-based approaches.